Question

I’m looking for a general strategy/advice on how to handle invalid UTF-8 input from users.

Even though my web application uses UTF-8, somehow some users enter invalid characters. This causes errors in PHP’s json_encode() and overall seems like a bad idea to have around.

W3C I18N FAQ: Multilingual Forms says "If non-UTF-8 data is received, an error message should be sent back.".

• How exactly should this be practically done, throughout a site with dozens of different places where data can be input?
• How do you present the error in a helpful way to the user?
• How do you temporarily store and display bad form data so the user doesn’t lose all their text? Strip bad characters? Use a replacement character, and how?
• For existing data in the database, when invalid UTF-8 data is detected, should I try to convert it and save it back (how? utf8_encode()? mb_convert_encoding()?), or leave as-is in the database but doing something (what?) before json_encode()?

I’m very familiar with the mbstring extension and am not asking "how does UTF-8 work in PHP?". I’d like advice from people with experience in real-world situations how they’ve handled this.

As part of the solution, I’d really like to see a fast method to convert invalid characters to U+FFFD.

Answer 1

The accept-charset="UTF-8" attribute is only a guideline for browsers to follow, and they are not forced to submit that in that way. Crappy form submission bots are a good example…

I usually ignore bad characters, either via iconv() or with the less reliable utf8_encode() / utf8_decode() functions. If you use iconv, you also have the option to transliterate bad characters.

Here is an example using iconv():

$str_ignore = iconv('UTF-8', 'UTF-8//IGNORE', $str);
$str_translit = iconv('UTF-8', 'UTF-8//TRANSLIT', $str);

If you want to display an error message to your users I’d probably do this in a global way instead of a per value received basis. Something like this would probably do just fine:

function utf8_clean($str)
{
    return iconv('UTF-8', 'UTF-8//IGNORE', $str);
}

$clean_GET = array_map('utf8_clean', $_GET);

if (serialize($_GET) != serialize($clean_GET))
{
    $_GET = $clean_GET;
    $error_msg = 'Your data is not valid UTF-8 and has been stripped.';
}

// $_GET is clean!

You may also want to normalize new lines and strip (non-)visible control chars, like this:

function Clean($string, $control = true)
{
    $string = iconv('UTF-8', 'UTF-8//IGNORE', $string);

    if ($control === true)
    {
            return preg_replace('~\p{C}+~u', '', $string);
    }

    return preg_replace(array('~\r\n?~', '~[^\P{C}\t\n]+~u'), array("\n", ''), $string);
}

---

Code to convert from UTF-8 to Unicode code points:

function Codepoint($char)
{
    $result = null;
    $codepoint = unpack('N', iconv('UTF-8', 'UCS-4BE', $char));

    if (is_array($codepoint) && array_key_exists(1, $codepoint))
    {
        $result = sprintf('U+%04X', $codepoint[1]);
    }

    return $result;
}

echo Codepoint('à'); // U+00E0
echo Codepoint('ひ'); // U+3072

It is probably faster than any other alternative, but I haven’t tested it extensively though.

---

Example:

$string = 'hello world�';

// U+FFFEhello worldU+FFFD
echo preg_replace_callback('/[\p{So}\p{Cf}\p{Co}\p{Cs}\p{Cn}]/u', 'Bad_Codepoint', $string);

function Bad_Codepoint($string)
{
    $result = array();

    foreach ((array) $string as $char)
    {
        $codepoint = unpack('N', iconv('UTF-8', 'UCS-4BE', $char));

        if (is_array($codepoint) && array_key_exists(1, $codepoint))
        {
            $result[] = sprintf('U+%04X', $codepoint[1]);
        }
    }

    return implode('', $result);
}

This may be what you were looking for.